Professional Documents
Culture Documents
(Ortiz Et Al) - A Vision System For An Underwater Cable Tracker
(Ortiz Et Al) - A Vision System For An Underwater Cable Tracker
(Ortiz Et Al) - A Vision System For An Underwater Cable Tracker
Mathematics and Computer Science Department, University of the Balearic Islands, 07071 Palma de Mallorca, Spain; e-mail: alberto.ortiz@clust.uib.es
Mathematics and Computer Science Department, University of the Balearic Islands, 07071 Palma de Mallorca, Spain
Mathematics and Computer Science Department, University of the Balearic Islands, 07071 Palma de Mallorca, Spain; e-mail: goliver@clust.uib.es
Abstract. Nowadays, the surveillance and inspection of underwater installations, such as power and telecommunication
cables and pipelines, is carried out by operators that, being
on the surface, drive a remotely operated vehicle (ROV) with
cameras mounted over it. This is a tedious and high timeconsuming task, easily prone to errors mainly because of
loss of attention or fatigue of the human operator. Besides,
the complexity of the task is increased by the lack of quality of typical seabed images, which are mainly characterised
by blurring, non-uniform illumination, lack of contrast and
instability in the vehicle motion. In this study, the development of a vision system guiding an autonomous underwater
vehicle (AUV) able to detect and track automatically an underwater power cable laid on the seabed is the main concern.
The vision system that is proposed tracks the cable with an
average success rate above 90%. The system has been tested
using sequences coming from a video tape obtained in several tracking sessions of various real cables with a ROV
driven from the surface. These cables were installed several
years ago, so that the images do not present highly contrasted cables over a sandy seabed; on the contrary, these
cables are partially covered in algae or sand, and are surrounded by other algae and rocks, thus making the sequences
highly realistic.
Key words: Features detection and tracking Image sequences Autonomous underwater vehicles Pipeline inspection
1 Introduction
The feasibility of an underwater installation consisting of
either cables, for power or telecommunication, or pipelines,
for gas or petrol, can only be guaranteed by means of an adequate inspection programme. This programme has to provide
the company with prompt information about potential hazardous situations or damages caused by the mobility of the
This study has been partially supported by the GOVERN BALEAR
(BOCAIB-16,3/2/98) and project CICYT-MAR99-1062-C03-03.
Correspondence to: A. Ortiz
130
2 Previous work
In the literature about cable inspection, two main sensing devices can be distinguished: the magnetometer and the sonar
(see [4, 7, 8], among others). The former detects the variations in the earth magnetic field induced by electrical and
telecommunication cables, even when they are buried. Using this device, however, requires isolating those parts of
the vehicle that generate additional magnetic fields so as
not to disturb the measurements of the magnetometer. The
other main sensing device, the multibeam sonar system, is
mainly used in exhaustive searching missions and to detect free-spans. In those cases, the main problem consists
in keeping the vehicle stabilized to improve the signal-tonoise ratio. Other systems that have also been used in cable
and pipelines inspection include seismic bottom profilers [4],
which combined with magnetometers allow getting a cross
section of the upper bottom sediments, and PIGS [18], vehicles of small dimensions that are introduced inside the
pipeline and use the present flow to move along.
In general, all these strategies need AUVs larger and
more powerful than is required by the mission because of
the size of the sensing devices and the consequent extra
batteries [5]. By using CCD cameras, however, this problem
is considerably reduced, both in cost and in AUV size. In
fact, in recent years several research groups have shown the
suitability of vision systems either for both navigation and
mission tasks (see [10, 16, 19, 20], among others).
With regard to visual cable and pipeline tracking and
inspection, several systems have been proposed so far. Matsumoto and Ito [9] developed a vision system able to follow electrical cables in underwater environments by using
edge detectors, the Hough transform and some higher-level
processing related to the line-like appearance of the cables.
Hallset [5] presented another system able to follow pipelines
also using edge detectors and the Hough transform, as well
as a map of the pipeline network. At the University of Ancona a system oriented towards helping human operators
in the inspection of gas and oil pipelines was also implemented [21]. In this case, the system detected the pipes and
some other accessories attached to them using statistical information obtained from selected areas of the image related
to the position of the cable. More recently, Balasuriya et al.
131
c
Image
ACQUISITION
NonValidIMG < N
Reset
KALMAN filter
Image
SEGMENTATION
Widen ROI
to the whole IMAGE
Cable
DETECTION
NonValidIMG = N
NonValidIMG =
NonValidIMG + 1
No OK
Cable detection
VALIDATION
OK
No OK
Prediction vs Measure
COMPARISON
OK
NonValidIMG = 0
ROI
PREDICTION
132
900
b
gray-level
number of pixels
800
700
600
500
400
300
200
100
obj. 2
gradient modulus
obj. 1
obj. 3
0
5
10
contour
zones
15
20
gradient
220
200
180
160
140
120
100
80
60
40
20
graylevel intensity
Fig. 3. a Ideal bidimensional histogram; b bidimensional histogram of a real image (Fig. 1a)
3500
450
400
3000
350
300
2500
250
200
2000
150
100
1500
50
0
0
1000
10
20
500
30
0
40
60
80
100
120
140
160
180
200
220
40
220
200
180
160
140
120
100
80
60
Fig. 4. Comparison between the unidimensional and the bidimensional histogram of a real image (Fig. 1d)
133
Orientation
FILTER
Segment
GROUPING
Long
straight
segments
Str. segments
FITTING
Filtered
straight
segments
Lines
EXTRACTION
Straight
segments
Contour
EXTRACTION
Lines
Segmented
IMAGE
Contours
Fig. 5. Segmentation in three classes for all the images in Fig. 1 (top row shows minimal-spanning tree results, and lower row shows K-means results)
Segment
SELECTION
Cable
PARAMETERS
3 1 2
5 (i, j) 4
(1)
0 0 0
The numbers indicate preference, where the lower the number, the higher the preference, except for zero, which represents a forbidden selection. Therefore, lines in vertical directions are favoured against horizontal or curving lines, according to the above-mentioned assumption. When, for a
given contour pixel, there is no adjacent pixel in the preferred orientations, the process of tracking the line finishes
134
w
S1
LSi
L1
L2
S
L
S2
Si
(2)
To compute the process noises vR and vL , several real sequences were manually analysed and the differences between
consecutive frames in the cable positions and orientations
were computed. At the end of this procedure, an estimation
of the covariance matrix of both noises was available. As for
the measurement noises wR and wL , the system was faced
against noisy synthetic sequences and the deviations between
the real orientations and positions of the cable and the measured ones were determined. The corresponding covariance
matrixes were finally obtained from those deviations.
Finally, the ROI for the next image is computed as it is
indicated in Eq. (3), where rL and rR represent the predicted
straight sides of the cable, and kL and kR are tolerance factors included in the ROI. Best results have been obtained by
135
Fig. 9. Intermediate and final results for image Fig. 1b: a segmented image; b contour image; c straight segments after the orientation filter; d detected sides
of the cable
Fig. 10. Intermediate and final results for image Fig. 1d: (a) segmented image; (b) contour image; c straight segments after the orientation filter; (d) detected
sides of the cable
computing them as twice the difference between the computed and the predicted position of each side.
ROI(t + 1) = [rL (t + 1) kL (t + 1), rR (t + 1) + kR (t + 1)](3)
8 Optimisations of the proposed method
The proposed method of detection and tracking the cable in
a sequence of images is highly time consuming. In order to
achieve a real-time performance, some optimizations have
been applied to the algorithm. They are detailed as follows:
First, among the several gradient operators proposed in
the computer vision literature e.g. forward differences,
centred finite differences, Roberts, Sobel, Prewitt, and
Canny (see [15, 17] for a discussion on the subject)
the one which has given the best results, both as a digital approximation to the real gradient of a gray-level
image and regarding computation time, has been the Sobel operator. As it is an operator involving weights that
are power of 2, it can be programmed in a very efficient
way using integer computations, which meaningfully accelerates the execution.
Second, as the clustering stage is, in the worst case, an
O(n3 ) process, where n is the number of initial classes
to group, diminishing n as much as possible dramatically contributes to reducing the computation time. After some experimental work, two optimisation strategies
have been introduced: (1) reduce the resolution of the
histogram by a factor kh ; and (2) reduce the classes to
group to a subset consisting of the kc most meaningful
ones, the ones representing the highest number of pixels.
The best results have been obtained for kh = 8 and selecting the kc = 20 classes representing the largest number of
pixels. On the one hand, this means that the limits of the
gray-level axis of the histogram are 0 and 256
8 1 = 31,
Fig. 11. Final results for two images of high complexity. Both pictures
show the detected sides of the cable
136
Fig. 12. Results for an excerpt of sequence 1. In the left image, the ROI appears superimposed together with the reconstructed sides of the cable (the
black lines correspond to the predicted sides). In the right image, the white line represents the axis of the cable and thus a possible command to the AUV
controller
9 System configuration
Fig. 13. Results for excerpts of sequences 2 (upper row left), 3 (upper row right), 4 (lower row left) and 5 (lower row right)
137
138
Detection
Tracking
Symbol
kh
kc
kf
ke
ks
kL
kR
Meaning
Histogram resolution reduction
Most important classes to be grouped
Number of final classes
Fitting error
Co-linearity error
Limit in the number of segments and straight lines
ROI left-side tolerance factor
ROI right-side tolerance factor
Typical value
8
20
3
2 image units
5 image units
100
2 |x L xL | image units
2 |x R xR | image units
Length (frames)
248
499
409
172
129
1457
Wrong detections
21
20
69
2
3
115
Success rate
92%
96%
83%
99%
98%
92%
References
1. Balasuriya B, Takai M, Lam W, Ura T, Kuroda Y (1997) Vision based
autonomous underwater vehicle navigation: underwater cable tracking.
In: Oceans 97 MTS/IEEE Conference Proceeding, vol 2, pp 1418
1424
2. Balasuriya B, Ura T (1999) Multi-sensor fusion for autonomous underwater cable tracking. In: Oceans 99 MTS/IEEE Riding the Crest
of the 21st Century, vol 1, pp 209215
3. Duda RO, Hart PE (1973) Pattern Classification and Scene Analysis.
Wiley, New York
4. Egeskov P, Bech M, Bowley R, Aage C (1995) Pipeline inspection
using an autonomous underwater vehicle. In: Proceedings of the 14th
International Conference on Offshore Mechanics and Arctic Engineering, vol 5, pp 539546
5. Hallset JO (1996) Testing the robustness of an underwater vision system. In: Laplante P, Stoyenko A (eds) Real-time imaging: theory, techniques, and applications. IEEE Press, pp 225260
6. Hartigan J (1975) Clustering algorithms. Wiley, New York
7. Ito Y, Kato N, Kojima J, Takagi S, Asakawa K, Shirasaki Y (1994)
Cable tracking for autonomous underwater vehicle. In: Proceedings of
the 1994 Symposium on Autonomous Underwater Vehicle Technology,
pp 218224
8. Iwanowski MD (1994) Surveillance unmanned underwater vehicle. In:
Proceedings of IEEE Oceans, pp I116I119
9. Matsumoto S, Ito Y (1995) Real-time vision-based tracking of submarine cables for AUV/ROV. In: Proceedings of MTS/IEEE Oceans, pp
19972002 Part 3 of 3
10. Nguyen HG, Kaomea PK, Heckman PJ Jr (1988) Machine visual guidance for an autonomous undersea submersible. In: Proceedings SPIE
Underwater Imaging, vol 980, pp 8289
139
11. Ortiz A, Oliver G, Frau J (1997) A vision system for underwater realtime control tasks. In: Oceans 97 MTS/IEEE Conference Proceeding,
vol 2, pp 14251430
12. Ortiz A, Simo M, Oliver G (2000) Real-time image sequence analysis for tracking an underwater power cable. Technical Report 2/2000,
Departament de Matem`atiques i Inform`atica (Universitat de les Illes
Balears)
13. Pal NR, Pal SK (1993) A review on image segmentation techniques.
Pattern Recogn 26(9):12771294
14. Panda DP, Rosenfeld A (1978) Image segmentation by pixel classification in (gray level,edge value) espace. IEEE Trans Comput C27(9):875879
15. Parker J (1997) Algorithms for image processing and computer vision.
Wiley, New York
16. Santos-Victor J, Sentieiro J (1994) The role of vision for underwater
vehicles. In: Proceedings of the IEEE Symposium on AUV Technology, pp 2835
17. Schalkoff RJ (1989) Digital image processing and computer vision.
Wiley, New York
18. Wearden T (1992) Transducers and sensors for pipeline inspection vehicles. Control Syst J 9(8):
19. Xu X, Negahdaripour S (1997) Vision-based motion sensing for underwater navigation and mosaicing of ocean floor images. In: Oceans
97 MTS/IEEE Conference Proceeding, vol 2, pp 14121417
20. Xu X, and Negahdaripour S (1999) Automatic optical station keeping and navigation of a ROV; sea trial experiments. In: Oceans 99
MTS/IEEE Riding the Crest of the 21st Century, vol 1, pp 7176
21. Zingaretti P, Tascini G, Puliti P, Zanoli S (1996) Imaging approach to
real-time tracking of submarine pipeline. In: Proceedings SPIE Electronic Imaging, vol 2661, pp 129137
Alberto Ortiz received his degree in Computer Science from the University of the Balearic
Islands (UIB) in 1992. He joined the Mathematics and Computer Science Department of
the UIB in 1992 and has been a faculty member since then. He is author and co-author of
several international publications related with
computer vision and robotics and has participated in several projects related with VLSI circuits design, fault-tolerant systems, and computer vision and robotics. Now, he is a member of the Systems, Robotics and Vision (SRV)
group of the UIB. His present research interests include land and underwater computer vision, land and underwater
robotics and real-time systems.
Miquel Simo received his degree in Computer Engineering from the University of the
Balearic Islands in 2000. He has participated
in projects related with underwater computer
vision and robotics. He is now a Software Engineer.
140